Has anyone ever built LAPS with gfortran? It appears that there is an attempt within the code (machine_dependent.F) to deal with the use of non-standard intrinsics (e.g. sind, cosd, etc), but gfortran still fails to find those "machine dependent" functions.

Another issue has come up compiling with gfortran. This is in src_tmatrix.f90. gfortran does not support the syntax 'real*16' for declaring a quad-precision real. I further found that when I tried to use the slightly-more-standard 'real(kind=16)' that it said Kind 16 is not supported. I think Intel (and AMD) hardware don't do 128-bit arithmetic, and gfortran doesn't do it in software either. I think 80-bit is the maximum that the Intel-ish hardware can do. This is confirmed by the fact that real(kind=10) compiles without error.

The 'really F90+ standard way' would be to call the intrinsic selected_real_kind with arguments of desired decimal precision and/or exponent range to get the appropriate kind parameter for the given hardware. So, my question is: How much precision is needed in this module?

Yes. It was happening in lga. There were issues (e.g. seg faults) in other executables, but I don't recall what they were now.

We now have 'gfortran' installed here in the LAPS development area so we set up a regular build. We are in the process of addressing some compile time issues. Once this compiles we can do runtime tests with selected programs.

Our compilation results can be seen here ('gfortran' entries):

http://laps.noaa.gov/builds/laps_builds.html

The 'gfortran' issue in the main LAPS library has been fixed so this should hopefully compile OK with our next weekly build.